Collegial Influence and Judicial Voting Change: The Effect of Membership Change on U.S. Supreme Court Justices

Transcription

1 Collegial Influence and Judicial Voting Change: The Effect of Membership Change on U.S. Supreme Court Justices 909 Scott R. Meinke Kevin M. Scott Understanding the source of voting changes by appellate judges provides an important window into the factors that shape the votes of the judges more generally. We argue that membership changes, by altering the collegial context in which judges make their choices, affect the information environment, long-term collegial considerations, and short-term strategic calculations. As a result, membership change should lead to greater uncertainty and more frequent voting changes among continuing justices in the term following a replacement. We test this proposition by looking at vote change by justices of the U.S. Supreme Court in two separate analyses: justices votes on search-andseizure cases since Mapp v. Ohio (1961) and on the progeny of Miranda v. Arizona (1966). Our results support the argument that the collegial context helps explain changes in voting choices. Our analysis suggests that collegial considerations are an important component of judges behavior and merit further evaluation in a cross-national context. Theoretical perspectives on judicial behavior provide good reason to expect that collegial court judgesfacross courts, nations, and levels of the judiciaryfare subject to intracourt influence. Judicial scholars know that, despite the importance of ideology, the collegial context in which judges decide cases has a significant effect on how their preferences are expressed. On rare occasions, judges can be quite forthright about the influence their colleagues have on their behavior, offering important insights into what factors influence their behavior. Tom Clark, a U.S. Supreme Court justice from 1949 to 1967, offered such an insight into his decisions leading up to the Court s landmark Mapp v. Ohio (1961) decision. Just before Clark became a justice, the Court decided Wolf v. Colorado (1949). This article reflects the views of the authors and does not necessarily reflect the views of the Congressional Research Service or the Library of Congress. For helpful comments and conversations on this work, we thank Lawrence Baum, Lauren Bell, Eileen Braman, Corey Ditslear, Rich Pacelle, Jeff Staton, and Margaret Williams, and we thank Carl Marchioli for excellent research assistance. Portions of this research were presented at the 2005 American Political Science Association annual meeting. Authors are listed alphabetically. Please direct correspondence to Scott Meinke, Department of Political Science, Bucknell University, Lewisburg, PA 17837; smeinke@bucknell.edu. Law & Society Review, Volume 41, Number 4 (2007) r 2007 by The Law and Society Association. All rights reserved.

2 910 Collegial Influence and Judicial Voting Change In that case, a six-justice majority declined to extend the exclusionary rule (suppressing evidence obtained in an illegal search) to the states. Through the 1950s, Court majorities held to this position, and Justice Clark joined these majorities. But in 1961, Justice Clark wrote the majority opinion in Mapp v. Ohio, reversing both Wolf and his own earlier support of that precedent. Justice Clark s writings help explain this interesting reversal. In Irvine v. California (1954), which upheld Wolf, Clark offered the following observations in a concurring opinion: Had I been here in 1949 when Wolf was decided, I would have applied the doctrine of Weeks v. United States, 232 U.S. 383 (1914), to the states. But the Court refused to do so then, and it still refuses today.... [I]t is with great reluctance that I follow Wolf. Perhaps strict adherence to the tenor of that decision may produce needed converts for its extinction (Irvine v. California, 347 U.S. 128, 138 [1954]). Justice Clark s unusually candid reflections nicely illustrate the quandary that collegial court judges face: their ability to affect decisions rests not only on their own preferences but also on their ability to convince their fellow judges of the merits of their position. In Justice Clark s case, building a majority required both the appointment of new justices allied with his sincere preference and the conversion of continuing justices (Schubert 1962:91 2). The experience of Justice Clark on the U.S. Supreme Court has been buttressed by scholarly findings in a variety of environments. For example, several scholars have concluded that under some circumstances, female judges behave differently from male judges (see, e.g., Allen & Wall 1987; Songer et al. 1994), and the presence of female judges may even change the behavior of male judges (O Connor & Segal 1990, but see Palmer 2002). Judges are concerned about the esteem of their colleagues (Baum 2006) and may alter their behavior in order to secure respect. Judges are also frequently characterized as strategic actors; if they are concerned about the ultimate content of policy, they may take account of the anticipated behavior of their colleagues. This literature is particularly well-developed on the U.S. courts of appeals, where regularly rotating panels for decisions allow the measurement of the impact of panel composition on judges decisionmaking behavior (Cross & Tiller 1998; Hettinger et al. 2004; Revesz 1997; Van Winkle 1996). In very different fields of research, scholars have produced considerable empirical evidence to highlight the impact of collegial context on judges individual choices, and all of this work suggests that collegial context matters to judges. If judicial decisionmaking is indeed subject to collegial influence, and if that influence is strong, then we should be able to

3 Meinke & Scott 911 witness changes in the voting positions of continuing judges when the composition of a collegial court changes. Although existing perspectives on judicial choice emphasize stability over time in individual judges choices, substantial intracourt influence should set the stage for judges to reconsider their choices when they find themselves in a significantly altered collegial environment: membership changes present new sources of information and persuasion, new prospects for building both working relationships with and respect from court colleagues, new opportunities for strategic short-term coalition-building, and new uncertainty over the relationship between choices and outcomes. To explore the systematic relationship between membership change and vote switching, we examine merits-vote choices on several recurring issues before the U.S. Supreme Court, looking for the effects of membership changes on continuing justices. To do so, we present a new approach to analyzing judicial decisionmaking over time, one that allows us to identify and assess vote switching by individual justices on similar questions over a relatively long period. Where some existing research has described the trajectory of appellate judges positions aggregated by term over a broad issue area (Baum 1992; Epstein, Hoekstra, et al. 1998; Ostberg et al. 2003), we compare the direction of individual votes from one case to the next. Our focus is on two enduring questions before the U.S. Supreme Court: the admissibility of searches under the Fourth Amendment after Mapp v. Ohio (1961), and the admissibility of confessions after Miranda v. Arizona (1966). As we explain below, this approach allows us to study the effects of membership changes on continuing justices choices because we can relatively easily isolate other factors that we know affect voting. The case-vote analysis also permits us to explore how case-level variables such as opinion writing relate to the stability of justices positions. Using this method, we find evidence that changes in the Supreme Court s composition make continuing justices more likely to reverse their positions on merits votes. While we do not disentangle the several theoretical origins of this change, we emphasize that the evidence reflects collegial influence broadly defined (see Revesz 1997: for similar conclusions on U.S. courts of appeals panels). The evidence comports with the role of compromise, deliberation, and consensus in collegial decisionmakingf in other words, changes in the intracourt context bring greater instability as justices are affected by new collegial interactions (Baum 1997:113 5). The findings likely also reflect the role of short-term strategic accounts, in which justices become more likely to change positions as they attempt to build new majorities or preserve outcomes in changing contexts. In presenting and discussing the findings, we highlight the connection between our

4 912 Collegial Influence and Judicial Voting Change U.S. Supreme Court findings and expectations about other collegial courts, and in the concluding section, we offer some specific discussion of the potential effect of institutional variables on the dynamics we describe. Voting Change on Collegial Courts Changes in a judge s behavior appear, at first glance, to be difficult to reconcile with the belief that high court judges in different countries and different institutional contexts seek to pursue their policy preferences (see, e.g., Ostberg et al. 2002). Most models of judicial behavior begin with the assumption that judges preferences are fixed and that their votes in cases are responses to the issues that each case presents. This assumption of stability, generally speaking, is not unreasonable. For judges, consistency is one of the objectives of decisionmaking (Baum 1997:26) and, for high court judges in particular, consistency allows lower courts and decision makers outside the judiciary to invest in future decisions on the assumption that judicial decisions in a policy area will remain relatively constant. 1 Furthermore, judges reliance on policy preferences for decisions suggests that their behavior should be consistent. If an attitude is a relatively enduring organization of beliefs about an object or situation (Rokeach 1968:134), then it follows that judges should have well-developed and consistent positions, and a judge s decision history itself could reinforce this stability. Finally, one might interpret the consistency of judicial behavior as an indicator of the degree to which judges make decisions free of external political pressure. Consistency in decisions may allow a court or its judges to demonstrate latitude (or lack thereof) to make and enforce decisions (Helmke 2002, 2005). This stability assumption has been made explicit in most of the major works on judicial behavior from both attitudinal and rational choice schools of thought, but it has been particularly important to the attitudinal model (Segal & Spaeth 2002). The attitudinal view is shaped by the work of Schubert (1962), who characterized the preferences of U.S. Supreme Court justices as representing independent ideal points (points on a continuum where individuals most prefer to locate public policy) and cases representing external 1 In the context of the U.S. Supreme Court and other common law courts, judges attach great importance to precedent (Epstein & Knight 1996; Spriggs & Hansford 2001), even if they do not adhere to it so closely that they allow precedent to supplant their policy preferences. Spaeth and Segal (1999) make much of the fact that justices refuse to alter their behavior to comply with a precedent with which they disagree. The view taken by Segal and Spaeth reinforces the argument that one should not expect justices to change their behavior, even if precedent suggests that they should do so (compare Richards & Kritzer 2002).

5 Meinke & Scott 913 stimuli to which the justices react. 2 Justices votes on the cases could be explained, in Schubert s view, by assessing the relationship between the stimuli of the cases and the location of the justices ideal points. Contemporary attitudinal work sustains the assumption of preference stability by relying on ideology measures that remain fixed over the justices careers (Segal & Cover 1989; Segal et al. 1995; Segal 1997), and within the strategic framework (Epstein & Knight 1998), the assumption of stable preferences is typically implicit, even though the logic of the strategic model anticipates that the expression of those preferences may not be stable over time. Sources of Voting Change To the extent that scholarship has considered behavioral change by judges, most inquiry has focused on sources of change in the aggregate policy outputs of courts (see Baum 1988, 1992). This collective voting change follows from a combination of three factors: membership change, issue change, and individual position change (Baum 1992:5). 3 While we understand these collective dynamics relatively well, our grasp of one of the collective change componentsfindividual changefis still fairly weak, although it has received some descriptive attention in recent analyses. For example, Epstein, Hoekstra, and colleagues (1998) show that there is observable change in the revealed preferences of U.S. Supreme Court justices, even when one accounts for issue change. While some justices move consistently in the conservative (Reed, White) or liberal (Blackmun, Clark) directions, other justices (Black, Douglas, Frankfurter, Powell, & Warren) exhibit somewhat more complex patterns. Ostberg et alia (2003:716), in a similar study of justices of the Supreme Court of Canada, found only two justices (Estey & Lamer) whose behavior later in their careers was significantly different from their earlier behavior once one accounts for issue change. The work of explaining individual change largely remains to be done. From a purely attitudinal perspective, explanation would begin and end with the influence of issue change and attitude change on individual judges. If individual judges change their behavior, either the judges preferences or the content of the cases (or both) have changed. And if judges have no reason to concern themselves with the preferences of their colleagues or with changes 2 Schubert and others have extended this work to other courts in other countries. Ostberg et alia (2002:236) and Smyth (2003) review the state of the literature for courts outside the United States. 3 In discussing the U.S. Supreme Court, Baum notes that individual-level position change may result from either attitude change or changes in other relevant forces, such as events external to the Court (1992:5).

6 914 Collegial Influence and Judicial Voting Change in the political environment, then the attitudinal perspective would provide a complete theoretical explanation for change. But if contextual factors do constrain judges, then issue and attitude change are not the only possible sources of change: shifts in the decisionmaking context, both internal and external, may also lead to individual-level change. The external context is altered by changing influences from external political actors, including other branches of the government, interest groups, and the media. The internal context of judges decisions, which is our focus here, is adjusted by changes in court membership. On a collegial court, the departure of one colleague and the arrival of a new one presents new choice considerations for continuing judges, particularly when the old and new judges contrast sharply in their attitudes. The effects of replacement may manifest themselves in several ways, all of which point toward a greater likelihood of position instability when court membership shifts. Membership change should affect the behavior of continuing judges, first, by increasing their uncertainty about how their positions in a particular case affect their broader, long-term goals. All collegial courts are characterized by repeated interactions among relatively autonomous actors, though the extent to which they interact may vary considerably by country, court, and era (see, e.g., Baum 2004:145; 2006:58 9; Narayan & Smyth 2005:150; Ostberg et al. 2003). These regular interactions mean that judges learn to cooperate and engage in reciprocity, rewarding those who have cooperated with them in the past and punishing others (Maltzman et al. 2000:20 1; see also Baum 1997:113; McCormick 2004:29; Murphy 1964:49 54). 4 Similarly, the collegial nature of high courts elevates the importance of consensus (Caldeira & Zorn 1998; Epstein, Segal, et al. 2001; Howard 1968; see, however, Allen & Wall 1987; Narayan & Smyth 2005; Smyth 2002) as well as the need for esteem from court colleagues (Baum 2006). The reinforcing pressures of consensus, reciprocity, and respect should give judges reason to reconsider how short-term choices will affect their long-term collegial relationships and their individual long-term goals. Similarly, collegial court judges value the respect of their intracourt colleagues and may adjust their behavior toward that goal (Baum 2006). More generally, since the collegial court environment inherently involves some deliberation and exchange of information (Kornhauser & Sager 1986:100 2), a change in personnel can expose continuing justices to new views and persuasive personalities. All of these long-term factors have the potential to bring new information and social considerations into judges 4 Engaging in this sort of generalized reciprocity is not, however, the same as constructing logrolls across specific cases.

7 Meinke & Scott 915 decisions, making membership change lead to position shifts on specific cases as new collegial considerations make past decisions less relevant as a guide to the current choice. Finally, replacing judges reconfigures the ideological alignment of the court, changing the short-term strategic calculations of the continuing judges. If judges ultimately seek to secure majorities for their preferred positions, they will compromise on doctrine (Epstein & Knight 1998; McCormick 2004) and may even change positions in order to influence the legal policy embodied in the majority opinion. 5 Membership change that alters the location of the median judge provides some judges new opportunities to build majorities, while others lose coveted bargaining positions at or near the median. The prospect of losing influence over the majority opinion may lead judges to conceal their sincere preferences if they know that expressing them will leave them in dissent and without any leverage on the majority. Similarly, membership change could alter the strategic situation in the opposite direction for an individual judge, depriving that judge of an incentive to behave strategically that was previously in place. In short, a change in ideological alignments should make a justice more likely to change his or her revealed preferences when a membership change occurs. 6 Evidence of Voting Change Empirical evidence demonstrates that judges on collegial courts make some short- and long-term behavioral changes, and some of this change may follow from changes in court composition. One example of this kind of voting change can be found in research on courts with rotating decision panels, where some evidence suggests short-term strategic behavior contingent on panel composition. If judges are strategically oriented toward securing policy outputs closest to their preferred outcomes, then judges who sit on courts that decide cases in rotating panels will exploit this institutional feature, behaving sincerely when their views hold a 5 On the U.S. Supreme Court, Chief Justice Warren Burger s well-documented manipulation of the opinion in Roe v. Wade (1973) is an extreme example of gaming the majority opinion (Maltzman et al. 2000:29). But significant movement may also occur from case to case in addition to occurring across different stages of a single case. 6 This view of judges as strategic actors on final votes contrasts with the sincere-actor perspective in some of the literature. For the U.S. Supreme Court, though, there is existing evidence that justices are not irrevocably attached to their sincere preferences. One piece of evidence comes from the existing literature on strategic decisionmaking, which highlights the fluidity of decisions between conference and final merits votes (Howard 1968; Maltzman & Wahlbeck 1996; Maltzman et al. 2000; Murphy 1964:56 68; but see Brenner 1980). It is also clear that justices bargain over opinion content, and this behavior demonstrates strategic flexibility in the service of policy goals (Epstein & Knight 1998:65 79, ).

8 916 Collegial Influence and Judicial Voting Change majority on the panel and strategically suppressing those views when in the panel minority in an effort to move the panel outcome toward their preferred policy position. There is an ongoing debate in the literature on the U.S. courts of appeals as to the likelihood of such behavior (Cross & Tiller 1998; Hettinger et al. 2004, 2006; Revesz 1997; Van Winkle 1996), and scholars of courts outside the United States have also found mixed results. Flemming (2004) suggests that such behavior by justices on leave panels of the Supreme Court of Canada is unlikely, given uncertainty about coram (panel) composition to hear appeals. His finding does not, however, rule out the possibility of strategic behavior at the merits stage. Given that Hausegger and Haynie find that the chief justices of the Supreme Court of Canada and the Appellate Division of the Supreme Court of the Republic of South Africa strategically assign justices who are ideologically closer to them to hear cases that involve politically salient issues (2003:655), one might expect strategic behavior by the justices on the panels as well. Existing work also provides mixed evidence for long-term behavioral shifts, although the sources of these shifts have received less empirical attention. As mentioned above, Epstein, Hoekstra, and colleagues (1998) establish that a considerable amount of longterm change exists in the aggregate voting behavior of individual justices on the U.S. Supreme Court, while Ostberg et alia (2003) find less long-term change by justices on the Supreme Court of Canada. 7 Relatedly, in his study of collective voting change on the U.S. Supreme Court, Baum (1992) found that collective change was not fully explained by the direct effects of replacement: for at least five of the eight natural court changes between 1949 and 1980, Baum attributes collective change, in part, to changes in the voting patterns of the continuing justices (1992:13). Baum concludes that the source of this collective change could not be divided between issue change and individual position change, but drawing on the congressional literature, he also observes that issue change may be controlled by focusing on relatively narrow areas of policy in which the subjects of legislative or judicial votes are relatively 7 Related research on the freshman effect suggests that U.S. Supreme Court justices exhibit higher variability in their votes early in their careers (and participate less in opinion writing) as they attempt to settle in to their career patterns (Hagle 1993; Howard 1968; Maltzman et al. 2000; Murphy 1964). Ostberg et alia (2003) investigate the freshman effect for justices of the Supreme Court of Canada. They argue that a freshman effect should be less likely on the Supreme Court of Canada due to a pervasive norm of consensus and a less-polarizing appointment process controlled by the prime minister, as opposed to the contentious process by which U.S. Supreme Court justices are nominated and confirmed. They find that there is no evidence of a freshman effectfvariability or change in voting patternsfbut that novice justices participate less in their first terms than they do later in their careers. Smyth (2005) finds that freshman justices on the High Court of Australia are less likely to dissent than their more senior colleagues.

9 Meinke & Scott 917 parallel in content (1992:6). Taken together, these existing findings demonstrate that voting change occurs, but they provide relatively little explanation for that change. Our general argument suggests that changing dynamics may explain at least some of the instability. The literature on court leadership and patterns of agreement provides some further evidence of changing judicial behavior. For example, scholars have observed long-term changes in behavior of associate justices who move to the chief justice chair on the U.S. Supreme Court (Lanier & Wood 2001). On the Supreme Court of Canada, Wetstein and Ostberg (2005) find that the promotion to chief justice increases the number of majority opinions a justice writes and decreases the number of dissenting votes (though, interestingly, save Chief Justice Beverley McLachlin, not the number of dissenting opinions authored). The effect may be particularly robust on the Supreme Court of Canada, given the chief justice s power to strike panels of different sizes. More relevant to our analysis, Ostberg and Wetstein find that membership change does not affect any of these patterns in chief justice behavior. Nonetheless, the move from associate to chief and the accompanying changes in behavior suggest that continuing justices can and do alter their behavior during their careers. Furthermore, some evidence suggests that membership change can have an impact on the degree to which the continuing justices agree or disagree with one another. McCormick (2004:28) has noted that changes in membership are often associated with changes in the frequency and patterns of dissent on the Supreme Court of Canada. During the Lamer Court, the departures of Justices John Sopinka (in 1997) and Gérard La Forest (in 1998) translated into lower dissent rates for four of the continuing justices (McCormick 2004:28). Membership change, then, affects both new justices and continuing justices on the Supreme Court of Canada, suggesting that continuing justices respond to the new opportunities created or eliminated by membership change. Smyth and Narayan (2004, see also Narayan & Smyth 2005; Smyth 2002, 2003) argue that weak leadership on the High Court of Australia has contributed to periods of high levels of dissent, and that membership changes (particularly those that create or eliminate personality conflicts) shape the behavior of continuing justices. Analytical Approach For our empirical analysis, we have focused on voting changes among continuing U.S. Supreme Court justices. In order to explore justices voting changes in relation to membership shifts, we need to draw not only theoretical but also empirical distinctions

10 918 Collegial Influence and Judicial Voting Change between the possible sources of individual vote change. These possible sourcesfattitude change, issue change, and membership changesfshould follow different patterns. Sincere preference change (attitude change) affects some justices significantly over time, but these patterns of change are individualizedfsome justices grow more liberal and some more conservative, while some exhibit little attitude change (Epstein, Hoekstra, et al. 1998). Issue change would affect the behavior of all justices similarly. If the Court s case mix grows harder 8 from one term to the next (due to membership change and its effect on the issue agenda), then one would expect conservative votes to be more likely for all of the justices. Voting change in response to membership change would be most likely to affect those justices whose preferred outcomes are made strategically more or less easy to achieve by the contextual shift; it should also affect other justices as they encounter new information in Court deliberations and as they experience more uncertainty in considering the link between their behavior and their long-term objectives. Thus, because content changes across cases can be controlled and because pure attitude shifts are not systematic across all justices (they do not all become more liberal or more conservative), we can associate other measurable behavior changes with change in the Court s internal alignments produced by membership change. This empirical approach could proceed at one of several levels of analysis. At the level of aggregate voting patterns, if issue change could be controlled successfully, we would expect to see significant shifts in justices level of liberal (conservative) voting in general issue areas. At the level of individual case votes, we would expect to see that individual justices become more (less) likely to support a liberal (conservative) outcome on similar cases over time when the intracourt strategic context shifts significantly. Both approaches have strengths and weaknesses. At the aggregate level, the most significant challenge is controlling successfully for issue change. While there are notable attempts to do so (Baum 1988, 1992, 1995; Epstein, Hoekstra, et al. 1998; Ostberg et al. 2003), individual case analysis facilitates fine-tuned controls for the change in the precise factual situations from case to case. Looking at individual cases also allows us to examine the possibility that stability at the aggregate level masks instability at the level of the individual case. At the level of the individual case, however, the justices may resist the appearance of changing. Accordingly, individual-case analysis may underestimate the true degree to which the justices 8 Harder and easier refer to the difficulty or ease of casting a liberal vote. The direction of this terminology throughout is not intended to convey a value judgment but rather to follow conventions in the literature that date at least to Schubert 1962.

11 Meinke & Scott 919 conceal their preferences and act strategically to build majorities on the Court. An additional disadvantage of this case-level approach may be generalizability. In order to account for this concern, we conduct two separate individual-level analyses of justices vote switches on recurring questions; the first examines search-andseizure cases since 1963, and the second studies the progeny of Miranda v. Arizona (1966). A focus on merits votes creates an additional limitation. The existing literature makes clear that justices care about the content of opinions and not simply about the direction of case outcomes, and the behavioral change we hypothesize should manifest itself not only in the merits votes justices cast but also in the content and tone of the opinions justices join and write. In this analysis, we assess the impact of membership change on the vote outcomes only, expecting that heightened instability in merits votes is a relatively tough test of the argument. Independent Variables Our argument posits that vote switching on recurring issues should be related to significant shifts in the Court s ideological composition. That is, when a new justice (or justices) joins the Court and causes the Court median to shift relative to the previous term, the continuing justices should reassess their positions. As we discussed above, behavioral changes in response to membership change may reflect collegial pressures (reciprocity, esteem), changes in information and deliberation, and shifts in the context for strategic behavior, but in each case, membership shifts with clear potential consequences for the Court s output are the ones that should induce the greatest response from continuing justices. While all of these changes in the decisionmaking environment may be brought about by the addition of any new justice, changes that shift the median of the Court (which we characterize as significant Court changes) should be most consequential in these instances since these replacements involve freshman justices who bring substantially different ideas to intracourt deliberation, significant new uncertainty about policy outcomes and long-term relationships, and new alignments for short-term strategic calculations. 9 The central hypothesis, then, is that a continuing justice will be more likely to change his or her vote on recurring issues in the first term of a new natural court that brings significant ideological movement, compared with other court terms. In the analyses, we operationalize significant natural court changes as those replacements that change the median aggregate 9 As we briefly examine herein, some of the theoretical sources of change would lead us to expect some vote change in the case of insignificant membership shifts as well.

12 920 Collegial Influence and Judicial Voting Change civil liberties support score of the Court. 10 For example, the median adjusted civil liberties score of the 1990 court termfthe last court of liberal justice Thurgood Marshall s tenurefwas If Justice Marshall is removed from this court and the first-term score of his replacement, conservative justice Clarence Thomas, is substituted, the 1990 court s median drops to This change indicates a significant court change in the conservative direction for the following term, Replacements that do not move the median of the Court are not coded as significant changes, even if there is an ideological contrast between the departing and arriving justices. During the time period covered by our analyses, significant median movements occurred in nine of the 14 terms with membership changes. 11 Since our exploration of the central hypothesis requires us to compare votes across cases, we must first address the comparability of the cases themselves. As we describe below, the cases in both datasets have been selected for their general comparability, but we know that on search-and-seizure (Segal 1984) and confessions cases (Benesh 2002; Benesh & Martinek 2002), variations in case facts have significant effects on justices choices. In order to account for this variation and to reveal the effects of other influences, we have created a series of fact variables for each of the two datasets. Relying on facts that previous research has established as salient in each area, we employ a series of indicator variables that summarize whether the changes in case facts from t 1tot make it easier or more difficult to support a pro-defendant position. We expect these indicators to be related to position changes in the following ways. First, as the fact pattern in a case makes excluding a confession or search easier than in the previous case, a vote change from pro-prosecution to pro-defendant should be more likely. Conversely, as the fact pattern in a case makes excluding a confession or search more difficult than in the previous case, a vote change from pro-defendant to pro-prosecution should be more likely. Details on the construction of the fact variables are outlined below as each analysis is introduced. 10 Aggregate support scores for this calculation are adjusted for issue change (Baum 1988). We use the adjusted civil liberties scores as an illustration here, as those scores are designed to account for issue change, meaning that over-time change in the scores reflects membership change and change by continuing justices. Other measures of Court tendencies (Martin & Quinn 2002) are more agnostic about what forces influence change, making them less appropriate in this particular context. 11 Terms with significant liberal changes include 1962, 1967, 1987, and Terms with significant conservative changes include 1969, 1970, 1971, 1975, and The indicator for significant court change equals 1 for each justice-vote cast during one of these terms with significant change (or, in a few instances, during the first term after the significant change in which a case in the issue area was decided).

13 Meinke & Scott 921 We account for other factors that can be expected to indicate a greater likelihood of position-switching in a given case. In particular, issuing a concurring opinion can be viewed as an outcomebased indicator of a justice s bargaining earlier in the process (Epstein & Knight 1998:76 9; Maltzman et al. 2000:66 9; Murphy 1964:54 68). Justices in the minority seeking votes for their position in the current case would be most likely to turn to a justice whom they perceive as a less-committed member of the majority, giving that justice bargaining leverage. It follows that the author of a (regular or special) concurring opinion at t 1 is more likely to change positions at t. If justices ultimately are looking to build majorities, then they may apply differential levels of pressure on their colleagues given the outcome the previous time an issue was considered. If the decision in the previous case was 7 2, the dissenters know they are three votes away from gaining a majority and may simply find it a waste of effort to convince other justices to switch. If, however, the previous decision was 5 4, then the dissenting justices know they are one vote switch away from the opportunity to hold a majority. Accordingly, they will be more willing to compromise with a justice in a minimum winning coalition (MWC) than with a justice in any other coalition. A justice who was a member of an MWC at t 1, then, should be more likely to change position at t. Characteristics of the individual justices may also affect their propensity to switch positions, and we account for this potential relationship in the analyses. Regardless of the case-specific factors described above, justices who are more moderate should be more likely to switch positions for several reasons. First, they are more likely to find themselves in a powerful bargaining position; they will find more opportunities to influence a majority by sacrificing some of their pure preferences. Similarly, it is possible that intangible costs of strategyfespecially the amount of ideological purity lost for the sake of influencefare higher for those justices farther from the median and lower for those at or near the median. Moderate justices can also be seen as more likely to respond, in their merits vote choices, to the influence of deliberation and persuasion among the justices. Finally, more moderate justices may observe more relevant differences from case to case: liberal Justice William Douglas may have been unlikely to distinguish between a search of a dorm room and an apartment, but moderate Justice Sandra Day O Connor may have found such a difference dispositive. For all these reasons, we expect that justices closer to the median will be more likely to change their votes, while those farther away will be less likely to do so, other things being equal. In the analyses, we define ideological location as justice j s absolute distance from the Court median (m) at time t. Our

14 922 Collegial Influence and Judicial Voting Change measures of ideological location are based on Martin and Quinn s aggregate estimates of justices ideal points (Martin & Quinn 2002); we take the absolute value of the difference between the ideal point of j and the ideal point of m. We approach the use of these estimates with a fair amount of caution, as using vote scores to predict votes may prove problematic due to problems of circularity. Using the scores as an independent variable in analysis of discrete issue areas, as we do here, raises far fewer concerns (Martin & Quinn 2005:5), so we believe this an appropriate and adequate measure of the justices ideological distances. A final factor, uncertainty, should affect the stability of all justices over time. In their study of switching from initial to final merits votes, Maltzman and Wahlbeck (1996) emphasize the role of uncertainty over vote choices in explaining instability. Among other factors, Maltzman and Wahlbeck find that the uncertainty related to lack of experience (freshman status) predicts switching prior to the final merits vote. In our longer-term view of vote switching, uncertainty should have a similar effect. Here, it is not simply tenure on the Court that we are interested in but also the length of a justice s voting history on a particular issue. 12 If uncertainty affects justices choices, then justices with shorter issuevoting histories to draw upon should reverse votes more often as they seek the position that best satisfies their goals within the collegial context. 13 In short, as a justice s voting history in an issue area grows longer, vote reversals from one case to the next should be less likely. 14 Methods Each analysis orders each justice s votes on all of the cases in chronological order and creates a variable that indicates whether a justice changed the direction of his or her vote (pro-defendant to pro-prosecution or vice versa) from one case to the next. Each set of cases is analyzed using a pooled binary time-series cross-sectional logit analysis in which the unit of analysis is the justice-case ( justice j s vote in the case at time t) and the dependent variable is the 12 Parallel research on legislative voting has demonstrated the importance of vote history in explaining stability and change in U.S. House vote choices. See Asher and Weisberg 1978 and Meinke The data in both analyses were shown to be duration-dependent in likelihood ratio tests, so the inclusion of a variable controlling for the duration of justice j in a stable voting position as of time t corrects for the effects of this duration dependence (see Beck et al on this general approach) as it also measures an effect of theoretical interest. 14 This control also allows us to account for vote switches that may appear idiosyncraticfunrelated to membership change. Justices with shorter vote histories should be more prone to such changes, while those with significant evidence of voting stability should be less likely to change for reasons unrelated to the included variables.

15 Meinke & Scott 923 indicator of vote change from t 1tot (1 5 changed vote, 0 5 same vote). 15 In order to account not only for position change itself but also for the direction of position change, we create a model that conditions the independent variables effects on the direction of the previous vote. Two versions of each independent variable are entered into the modelfone for justices with a pro-prosecution vote at t 1 and one for justices with a pro-defendant vote at t 1. Where a justice voted pro-prosecution at t 1, the pro-prosecution variables take on their actual values, and the pro-defendant variables are all set to zero. Similarly, for a justice with a t 1 prodefendant vote, the slate of pro-defendant variables reflects their actual values, while the pro-prosecution variables are zero. The model also includes an indicator variable for the justice s position at t 1. The coefficients for the two sets of independent variables, then, indicate the effect of each variable conditional on the justice s voting position at t 1. This approach to modeling conditional effects, which is described in detail by Wright (1976), yields results that are identical to a conventional dummy-variable interaction model. We follow this approach for the cleaner interpretation and presentation of conditional effects that it affords. For a recent application of this approach, see Giles et alia (2001). To account for the time-series cross-sectional nature of the data, we estimate a population-averaged logit. While conditional models are more useful when the primary question of interest is the effect of changes in covariates within a particular observation, [marginal models] are more valuable for making comparisons across groups or subpopulations (Zorn 2001:475). Using a marginal, or population-averaged (PA), model avoids estimating cluster-specific effects (the approach a fixed-effects model would take) or assuming that the cluster-specific effect follows a stochastic distribution. Rather, a PA model adjusts the covariance matrix of the estimated parameters to account for non-independence across observations or time points (Zorn 2001:474). As a result, coefficients in a PA model represent the average effect, across the entire population, of a one-unit shift in X it on Pr(Y it ) (Zorn 2001:474 5). This approach is desirable in our application because we are interested in understanding effects across subpopulations (here, the individual justices whose votes are assumed to be correlated). 15 The data for both analyses originally contained several cases in which decisions were announced on the same day. These cases are separate decisions with distinct U.S. Reports citations; in other words, they are not simply cases with differing docket numbers joined together for a decision (such cases are obviously treated in the dataset as one decision). In each analysis, we have included votes on these distinct but simultaneously decided cases as separate sets of observations in the longitudinal analysis, ordering them consistently by their appearance in the U.S. Reports. Eliminating all of the simultaneously decided cases except for the first on each date does not substantially change the results for either model.

16 924 Collegial Influence and Judicial Voting Change Search and Seizure To analyze vote change on U.S. Supreme Court search-andseizure cases, we begin with the dataset collected by Segal (Segal 1984, 1985; Segal & Spaeth 2002) and updated by Kritzer and Richards (Kritzer & Richards 2005). 16 The dataset includes every search-and-seizure case decided by the Supreme Court from its landmark decision in Mapp v. Ohio (1961) through the 2002 term. The issues presented in these cases do vary somewhat from case to case, requiring us to account for changing fact patterns. To establish the fact profile of each case relative to its predecessor, we create an indicator of intrusiveness, which we generate for each search. 17 Searches with positive values are more intrusive than searches with negative values. We then calculate the difference between the case at t 1 and t. If the value is negative, the search under consideration is less intrusive than the search under review in the previous case. A positive value means the search is more intrusive than the previous search. We then recode this to two dummy variables: more intrusive and less intrusive. More intrusive takes on the value 1 if the case at t represents a search that is more intrusive than the previous search, and is 0 otherwise. Less intrusive takes on the value 1 if the case at t represents a search that is less intrusive than the previous search, and is 0 otherwise. 18 Results Table 1 presents the results of the search-and-seizure analysis. As described above, the first set of coefficients represents the independent variables effects conditional on a pro-prosecution vote at t 1. In other words, the coefficients show the effect of each variable on the likelihood of a switch from a pro-prosecution to a 16 We thank Jeffrey Segal, Herbert Kritzer, and Mark Richards for sharing the data. 17 Intrusiveness is calculated using the following equation: (house) (business) (person) (car) (full search) (warrant issued) (lower court finding of probable cause) (incident to arrest) (after arrest) (after unlawful arrest) (exceptions). The weights are derived from a logistic regression of all of the justice-votes, and the regression includes Segal-Cover measures of the justices ideology. This approach follows the approach to measuring intrusiveness of a search taken by Cameron et alia 2000, though our equation includes probable cause. 18 One fundamental problem remains: several cases in the dataset involve the admissibility of more than one search. In some cases, there is more than one defendant. In other cases, there is more than one searchfsay, the search of a home and a vehicle. Collapsing the searches would overstate the intrusiveness and not accurately reflect the decisionmaking process, and some justices allow one search but not the other. Arbitrarily determining which search comes first in the justices sequence also seems to be inaccurate, so we exclude from our analysis the 19 cases (of a total 202 cases) where more than one search is considered by the justices.

17 Meinke & Scott 925 Table 1. Vote Changes on Search-and-Seizure Cases, b S.E. Pro-Prosecution Vote at t 1 Less Intrusive (0.395) More Intrusive n (0.379) Significant Natural Court Change n (0.199) Member of Minimum Winning Coalition t (0.202) Concurrence Author t nn (0.231) Distance From Court Median (0.061) Pro-Defendant Vote at t 1 Less Intrusive n (0.496) More Intrusive (0.507) Significant Natural Court Change nn (0.226) Member of Minimum Winning Coalition t (0.212) Concurrence Author t (0.266) Distance From Court Median nn (0.058) Stable Votes nn (0.019) Pro-Defendant Vote t (0.649) Constant n (0.383) Wald w 2 (14 d.f.) Pr4w 2 o0.001 Observations 1537 Cell entries are population-averaged logit coefficients. Standard errors in parentheses. n po0.05; nn po0.01; nnn po0.001, two-tailed tests. pro-defendant position. We find, first, that justices are more likely to change to a pro-defendant vote when the search is more intrusive. Looking at the second set of coefficients (for justices with prodefendant votes at t 1), we see that justices are more likely to change from a pro-defendant to a pro-prosecution vote when the search is less intrusive than the search considered in the previous case. These results provide an important validity check on the construction of the intrusiveness variable and confirm that the justices are sensitive to changes in the fact patterns. Consistent with our central hypothesis, justices are significantly more likely to change their vote positions in the term following a significant change in the Court s composition. The positive and statistically significant coefficients on the court-change variables in both sets of coefficients indicate that justices are more likely to defect from a pro-prosecution position and from a pro-defendant position. In short, the evidence indicates that the justices react to membership changes by becoming more likely to switch their positions in search-and-seizure cases. Justices who express doubt about the drift of the majority coalition by concurring in the previous case are, in fact, more likely to change their votes in the subsequent case, but this finding holds only for justices who voted pro-prosecution in the previous case. Justices who cast pro-defendant votes in the previous case and concurred are neither more nor less likely to change their votes. Justices who are members of MWCs are no more likely to change votes than justices who cast votes in cases that are less narrowly